kernel - Major signal path adjustments to fix races, tsleep race fixes, +more
[dragonfly.git] / sys / kern / kern_exec.c
CommitLineData
984263bc
MD
1/*
2 * Copyright (c) 1993, David Greenman
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 *
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24 * SUCH DAMAGE.
25 *
26 * $FreeBSD: src/sys/kern/kern_exec.c,v 1.107.2.15 2002/07/30 15:40:46 nectar Exp $
27 */
28
29#include <sys/param.h>
30#include <sys/systm.h>
31#include <sys/sysproto.h>
32#include <sys/kernel.h>
33#include <sys/mount.h>
34#include <sys/filedesc.h>
35#include <sys/fcntl.h>
36#include <sys/acct.h>
37#include <sys/exec.h>
38#include <sys/imgact.h>
39#include <sys/imgact_elf.h>
2bd9d75c 40#include <sys/kern_syscall.h>
984263bc
MD
41#include <sys/wait.h>
42#include <sys/malloc.h>
43#include <sys/proc.h>
895c1f85 44#include <sys/priv.h>
29f58392 45#include <sys/ktrace.h>
984263bc
MD
46#include <sys/signalvar.h>
47#include <sys/pioctl.h>
fad57d0e 48#include <sys/nlookup.h>
984263bc
MD
49#include <sys/sysent.h>
50#include <sys/shm.h>
51#include <sys/sysctl.h>
52#include <sys/vnode.h>
a6f89c72 53#include <sys/vmmeter.h>
17a7ca1f 54#include <sys/libkern.h>
984263bc 55
5c5185ae
SG
56#include <cpu/lwbuf.h>
57
984263bc
MD
58#include <vm/vm.h>
59#include <vm/vm_param.h>
60#include <sys/lock.h>
61#include <vm/pmap.h>
62#include <vm/vm_page.h>
63#include <vm/vm_map.h>
64#include <vm/vm_kern.h>
65#include <vm/vm_extern.h>
66#include <vm/vm_object.h>
a55afca2 67#include <vm/vnode_pager.h>
984263bc
MD
68#include <vm/vm_pager.h>
69
70#include <sys/user.h>
527fddf7 71#include <sys/reg.h>
984263bc 72
19bfc8ab 73#include <sys/refcount.h>
5fd012e0 74#include <sys/thread2.h>
684a93c4 75#include <sys/mplock2.h>
5fd012e0 76
984263bc 77MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
18b3f457 78MALLOC_DEFINE(M_EXECARGS, "exec-args", "Exec arguments");
984263bc 79
402ed7e1 80static register_t *exec_copyout_strings (struct image_params *);
984263bc
MD
81
82/* XXX This should be vm_size_t. */
83static u_long ps_strings = PS_STRINGS;
84SYSCTL_ULONG(_kern, KERN_PS_STRINGS, ps_strings, CTLFLAG_RD, &ps_strings, 0, "");
85
86/* XXX This should be vm_size_t. */
87static u_long usrstack = USRSTACK;
88SYSCTL_ULONG(_kern, KERN_USRSTACK, usrstack, CTLFLAG_RD, &usrstack, 0, "");
89
90u_long ps_arg_cache_limit = PAGE_SIZE / 16;
91SYSCTL_LONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
92 &ps_arg_cache_limit, 0, "");
93
94int ps_argsopen = 1;
95SYSCTL_INT(_kern, OID_AUTO, ps_argsopen, CTLFLAG_RW, &ps_argsopen, 0, "");
96
5c627295
MD
97static int ktrace_suid = 0;
98SYSCTL_INT(_kern, OID_AUTO, ktrace_suid, CTLFLAG_RW, &ktrace_suid, 0, "");
99
2bd9d75c
DRJ
100void print_execve_args(struct image_args *args);
101int debug_execve_args = 0;
102SYSCTL_INT(_kern, OID_AUTO, debug_execve_args, CTLFLAG_RW, &debug_execve_args,
103 0, "");
104
17a7ca1f 105/*
18b3f457
MD
106 * Exec arguments object cache
107 */
108static struct objcache *exec_objcache;
109
110static
111void
112exec_objcache_init(void *arg __unused)
113{
0aa16b5d 114 int cluster_limit;
1f29f2c5
MD
115 size_t limsize;
116
117 /*
118 * Maximum number of concurrent execs. This can be limiting on
119 * systems with a lot of cpu cores but it also eats a significant
120 * amount of memory.
121 */
122 cluster_limit = 16;
123 limsize = kmem_lim_size();
124 if (limsize > 7 * 1024)
125 cluster_limit *= 2;
126 if (limsize > 15 * 1024)
127 cluster_limit *= 2;
0aa16b5d 128
18b3f457
MD
129 exec_objcache = objcache_create_mbacked(
130 M_EXECARGS, PATH_MAX + ARG_MAX,
765b1ae0 131 &cluster_limit, 8,
18b3f457
MD
132 NULL, NULL, NULL);
133}
134SYSINIT(exec_objcache, SI_BOOT2_MACHDEP, SI_ORDER_ANY, exec_objcache_init, 0);
135
136/*
17a7ca1f
MD
137 * stackgap_random specifies if the stackgap should have a random size added
138 * to it. It must be a power of 2. If non-zero, the stack gap will be
0ced1954 139 * calculated as: ALIGN(karc4random() & (stackgap_random - 1)).
17a7ca1f
MD
140 */
141static int stackgap_random = 1024;
142static int
143sysctl_kern_stackgap(SYSCTL_HANDLER_ARGS)
144{
145 int error, new_val;
146 new_val = stackgap_random;
147 error = sysctl_handle_int(oidp, &new_val, 0, req);
148 if (error != 0 || req->newptr == NULL)
149 return (error);
150 if ((new_val < 0) || (new_val > 16 * PAGE_SIZE) || ! powerof2(new_val))
151 return (EINVAL);
152 stackgap_random = new_val;
153
154 return(0);
155}
156
157SYSCTL_PROC(_kern, OID_AUTO, stackgap_random, CTLFLAG_RW|CTLTYPE_UINT,
158 0, 0, sysctl_kern_stackgap, "IU", "Max random stack gap (power of 2)");
159
2bd9d75c
DRJ
160void
161print_execve_args(struct image_args *args)
162{
163 char *cp;
164 int ndx;
165
166 cp = args->begin_argv;
167 for (ndx = 0; ndx < args->argc; ndx++) {
6ea70f76 168 kprintf("\targv[%d]: %s\n", ndx, cp);
2bd9d75c
DRJ
169 while (*cp++ != '\0');
170 }
171 for (ndx = 0; ndx < args->envc; ndx++) {
6ea70f76 172 kprintf("\tenvv[%d]: %s\n", ndx, cp);
2bd9d75c
DRJ
173 while (*cp++ != '\0');
174 }
175}
176
984263bc
MD
177/*
178 * Each of the items is a pointer to a `const struct execsw', hence the
179 * double pointer here.
180 */
181static const struct execsw **execsw;
182
5417cd57
SS
183/*
184 * Replace current vmspace with a new binary.
185 * Returns 0 on success, > 0 on recoverable error (use as errno).
90947cb1 186 * Returns -1 on lethal error which demands killing of the current
5417cd57
SS
187 * process!
188 */
984263bc 189int
fad57d0e 190kern_execve(struct nlookupdata *nd, struct image_args *args)
984263bc 191{
dadab5e9 192 struct thread *td = curthread;
08f2f1bb 193 struct lwp *lp = td->td_lwp;
dadab5e9 194 struct proc *p = td->td_proc;
984263bc 195 register_t *stack_base;
19bfc8ab 196 struct pargs *pa;
6fa9e71a
MD
197 struct sigacts *ops;
198 struct sigacts *nps;
984263bc
MD
199 int error, len, i;
200 struct image_params image_params, *imgp;
201 struct vattr attr;
402ed7e1 202 int (*img_first) (struct image_params *);
984263bc 203
2bd9d75c 204 if (debug_execve_args) {
6ea70f76 205 kprintf("%s()\n", __func__);
2bd9d75c
DRJ
206 print_execve_args(args);
207 }
208
dadab5e9 209 KKASSERT(p);
19bfc8ab 210 lwkt_gettoken(&p->p_token);
984263bc
MD
211 imgp = &image_params;
212
213 /*
dc52e1cc
MD
214 * NOTE: P_INEXEC is handled by exec_new_vmspace() now. We make
215 * no modifications to the process at all until we get there.
216 *
217 * Note that multiple threads may be trying to exec at the same
218 * time. exec_new_vmspace() handles that too.
984263bc 219 */
984263bc
MD
220
221 /*
222 * Initialize part of the common data
223 */
224 imgp->proc = p;
2bd9d75c 225 imgp->args = args;
984263bc 226 imgp->attr = &attr;
984263bc 227 imgp->entry_addr = 0;
29802dbb 228 imgp->resident = 0;
984263bc
MD
229 imgp->vmspace_destroyed = 0;
230 imgp->interpreted = 0;
2abfe22f 231 imgp->interpreter_name[0] = 0;
984263bc
MD
232 imgp->auxargs = NULL;
233 imgp->vp = NULL;
234 imgp->firstpage = NULL;
235 imgp->ps_strings = 0;
adc42cf3
JM
236 imgp->execpath = imgp->freepath = NULL;
237 imgp->execpathp = 0;
18f40545 238 imgp->image_header = NULL;
2bd9d75c
DRJ
239
240interpret:
984263bc
MD
241
242 /*
fad57d0e
MD
243 * Translate the file name to a vnode. Unlock the cache entry to
244 * improve parallelism for programs exec'd in parallel.
984263bc 245 */
fad57d0e
MD
246 if ((error = nlookup(nd)) != 0)
247 goto exec_fail;
28623bf9 248 error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &imgp->vp);
fad57d0e
MD
249 KKASSERT(nd->nl_flags & NLC_NCPISLOCKED);
250 nd->nl_flags &= ~NLC_NCPISLOCKED;
28623bf9 251 cache_unlock(&nd->nl_nch);
fad57d0e 252 if (error)
984263bc 253 goto exec_fail;
984263bc
MD
254
255 /*
246693ac
AHJ
256 * Check file permissions (also 'opens' file).
257 * Include also the top level mount in the check.
984263bc 258 */
246693ac 259 error = exec_check_permissions(imgp, nd->nl_nch.mount);
984263bc 260 if (error) {
a11aaa81 261 vn_unlock(imgp->vp);
984263bc
MD
262 goto exec_fail_dealloc;
263 }
264
265 error = exec_map_first_page(imgp);
a11aaa81 266 vn_unlock(imgp->vp);
984263bc
MD
267 if (error)
268 goto exec_fail_dealloc;
269
315b8b8b
JM
270 imgp->proc->p_osrel = 0;
271
2bd9d75c 272 if (debug_execve_args && imgp->interpreted) {
6ea70f76
SW
273 kprintf(" target is interpreted -- recursive pass\n");
274 kprintf(" interpreter: %s\n", imgp->interpreter_name);
2bd9d75c
DRJ
275 print_execve_args(args);
276 }
277
984263bc
MD
278 /*
279 * If the current process has a special image activator it
280 * wants to try first, call it. For example, emulating shell
281 * scripts differently.
282 */
283 error = -1;
284 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
285 error = img_first(imgp);
286
287 /*
29802dbb
MD
288 * If the vnode has a registered vmspace, exec the vmspace
289 */
290 if (error == -1 && imgp->vp->v_resident) {
291 error = exec_resident_imgact(imgp);
292 }
293
294 /*
984263bc
MD
295 * Loop through the list of image activators, calling each one.
296 * An activator returns -1 if there is no match, 0 on success,
297 * and an error otherwise.
298 */
299 for (i = 0; error == -1 && execsw[i]; ++i) {
300 if (execsw[i]->ex_imgact == NULL ||
301 execsw[i]->ex_imgact == img_first) {
302 continue;
303 }
304 error = (*execsw[i]->ex_imgact)(imgp);
305 }
306
307 if (error) {
308 if (error == -1)
309 error = ENOEXEC;
310 goto exec_fail_dealloc;
311 }
312
313 /*
314 * Special interpreter operation, cleanup and loop up to try to
315 * activate the interpreter.
316 */
317 if (imgp->interpreted) {
318 exec_unmap_first_page(imgp);
fad57d0e
MD
319 nlookup_done(nd);
320 vrele(imgp->vp);
321 imgp->vp = NULL;
322 error = nlookup_init(nd, imgp->interpreter_name, UIO_SYSSPACE,
323 NLC_FOLLOW);
324 if (error)
325 goto exec_fail;
984263bc
MD
326 goto interpret;
327 }
328
329 /*
adc42cf3
JM
330 * Do the best to calculate the full path to the image file
331 */
332 if (imgp->auxargs != NULL &&
333 ((args->fname != NULL && args->fname[0] == '/') ||
334 vn_fullpath(imgp->proc,
335 imgp->vp,
336 &imgp->execpath,
337 &imgp->freepath,
338 0) != 0))
339 imgp->execpath = args->fname;
340
341 /*
984263bc
MD
342 * Copy out strings (args and env) and initialize stack base
343 */
344 stack_base = exec_copyout_strings(imgp);
345 p->p_vmspace->vm_minsaddr = (char *)stack_base;
346
347 /*
348 * If custom stack fixup routine present for this process
29802dbb
MD
349 * let it do the stack setup. If we are running a resident
350 * image there is no auxinfo or other image activator context
351 * so don't try to add fixups to the stack.
352 *
984263bc
MD
353 * Else stuff argument count as first item on stack
354 */
29802dbb 355 if (p->p_sysent->sv_fixup && imgp->resident == 0)
984263bc
MD
356 (*p->p_sysent->sv_fixup)(&stack_base, imgp);
357 else
2bd9d75c 358 suword(--stack_base, imgp->args->argc);
984263bc
MD
359
360 /*
361 * For security and other reasons, the file descriptor table cannot
362 * be shared after an exec.
363 */
364 if (p->p_fd->fd_refcnt > 1) {
365 struct filedesc *tmp;
366
2994659f
VS
367 error = fdcopy(p, &tmp);
368 if (error != 0)
369 goto exec_fail;
0a4a9c77 370 fdfree(p, tmp);
984263bc
MD
371 }
372
373 /*
374 * For security and other reasons, signal handlers cannot
375 * be shared after an exec. The new proces gets a copy of the old
376 * handlers. In execsigs(), the new process will have its signals
377 * reset.
378 */
6fa9e71a
MD
379 ops = p->p_sigacts;
380 if (ops->ps_refcnt > 1) {
381 nps = kmalloc(sizeof(*nps), M_SUBPROC, M_WAITOK);
382 bcopy(ops, nps, sizeof(*nps));
383 refcount_init(&nps->ps_refcnt, 1);
384 p->p_sigacts = nps;
385 if (refcount_release(&ops->ps_refcnt)) {
386 kfree(ops, M_SUBPROC);
387 ops = NULL;
388 }
984263bc
MD
389 }
390
0daa37a5
MD
391 /*
392 * For security and other reasons virtual kernels cannot be
393 * inherited by an exec. This also allows a virtual kernel
394 * to fork/exec unrelated applications.
395 */
4a22e893
MD
396 if (p->p_vkernel)
397 vkernel_exit(p);
0daa37a5 398
984263bc
MD
399 /* Stop profiling */
400 stopprofclock(p);
401
402 /* close files on exec */
403 fdcloseexec(p);
404
405 /* reset caught signals */
406 execsigs(p);
407
408 /* name this process - nameiexec(p, ndp) */
28623bf9
MD
409 len = min(nd->nl_nch.ncp->nc_nlen, MAXCOMLEN);
410 bcopy(nd->nl_nch.ncp->nc_name, p->p_comm, len);
984263bc 411 p->p_comm[len] = 0;
08f2f1bb 412 bcopy(p->p_comm, lp->lwp_thread->td_comm, MAXCOMLEN+1);
984263bc
MD
413
414 /*
415 * mark as execed, wakeup the process that vforked (if any) and tell
416 * it that it now has its own resources back
417 */
4643740a
MD
418 p->p_flags |= P_EXEC;
419 if (p->p_pptr && (p->p_flags & P_PPWAIT)) {
420 p->p_flags &= ~P_PPWAIT;
984263bc
MD
421 wakeup((caddr_t)p->p_pptr);
422 }
423
424 /*
425 * Implement image setuid/setgid.
426 *
427 * Don't honor setuid/setgid if the filesystem prohibits it or if
428 * the process is being traced.
429 */
430 if ((((attr.va_mode & VSUID) && p->p_ucred->cr_uid != attr.va_uid) ||
431 ((attr.va_mode & VSGID) && p->p_ucred->cr_gid != attr.va_gid)) &&
432 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
4643740a 433 (p->p_flags & P_TRACED) == 0) {
984263bc
MD
434 /*
435 * Turn off syscall tracing for set-id programs, except for
436 * root. Record any set-id flags first to make sure that
437 * we do not regain any tracing during a possible block.
438 */
41c20dac 439 setsugid();
5c627295
MD
440 if (p->p_tracenode && ktrace_suid == 0 &&
441 priv_check(td, PRIV_ROOT) != 0) {
29f58392
MD
442 ktrdestroy(&p->p_tracenode);
443 p->p_traceflag = 0;
984263bc
MD
444 }
445 /* Close any file descriptors 0..2 that reference procfs */
446 setugidsafety(p);
447 /* Make sure file descriptors 0..2 are in use. */
f3a2d8c4 448 error = fdcheckstd(lp);
984263bc
MD
449 if (error != 0)
450 goto exec_fail_dealloc;
451 /*
452 * Set the new credentials.
453 */
e9a372eb 454 cratom(&p->p_ucred);
984263bc 455 if (attr.va_mode & VSUID)
41c20dac 456 change_euid(attr.va_uid);
984263bc
MD
457 if (attr.va_mode & VSGID)
458 p->p_ucred->cr_gid = attr.va_gid;
98a7f915
MD
459
460 /*
461 * Clear local varsym variables
462 */
463 varsymset_clean(&p->p_varsymset);
984263bc 464 } else {
41c20dac
MD
465 if (p->p_ucred->cr_uid == p->p_ucred->cr_ruid &&
466 p->p_ucred->cr_gid == p->p_ucred->cr_rgid)
4643740a 467 p->p_flags &= ~P_SUGID;
984263bc
MD
468 }
469
470 /*
471 * Implement correct POSIX saved-id behavior.
472 */
115ccd83
MD
473 if (p->p_ucred->cr_svuid != p->p_ucred->cr_uid ||
474 p->p_ucred->cr_svgid != p->p_ucred->cr_gid) {
475 cratom(&p->p_ucred);
476 p->p_ucred->cr_svuid = p->p_ucred->cr_uid;
477 p->p_ucred->cr_svgid = p->p_ucred->cr_gid;
478 }
984263bc
MD
479
480 /*
481 * Store the vp for use in procfs
482 */
483 if (p->p_textvp) /* release old reference */
484 vrele(p->p_textvp);
fad57d0e
MD
485 p->p_textvp = imgp->vp;
486 vref(p->p_textvp);
984263bc 487
8ba5f7ef
AH
488 /* Release old namecache handle to text file */
489 if (p->p_textnch.ncp)
490 cache_drop(&p->p_textnch);
491
492 if (nd->nl_nch.mount)
493 cache_copy(&nd->nl_nch, &p->p_textnch);
494
984263bc
MD
495 /*
496 * Notify others that we exec'd, and clear the P_INEXEC flag
497 * as we're now a bona fide freshly-execed process.
498 */
499 KNOTE(&p->p_klist, NOTE_EXEC);
4643740a 500 p->p_flags &= ~P_INEXEC;
984263bc
MD
501
502 /*
503 * If tracing the process, trap to debugger so breakpoints
504 * can be set before the program executes.
505 */
506 STOPEVENT(p, S_EXEC, 0);
507
4643740a 508 if (p->p_flags & P_TRACED)
84204577 509 ksignal(p, SIGTRAP);
984263bc
MD
510
511 /* clear "fork but no exec" flag, as we _are_ execing */
512 p->p_acflag &= ~AFORK;
513
514 /* Set values passed into the program in registers. */
08f2f1bb 515 exec_setregs(imgp->entry_addr, (u_long)(uintptr_t)stack_base,
984263bc
MD
516 imgp->ps_strings);
517
349433c9
MD
518 /* Set the access time on the vnode */
519 vn_mark_atime(imgp->vp, td);
520
19bfc8ab
MD
521 /*
522 * Free any previous argument cache
523 */
524 pa = p->p_args;
984263bc 525 p->p_args = NULL;
19bfc8ab
MD
526 if (pa && refcount_release(&pa->ar_ref)) {
527 kfree(pa, M_PARGS);
528 pa = NULL;
529 }
984263bc 530
19bfc8ab
MD
531 /*
532 * Cache arguments if they fit inside our allowance
533 */
2bd9d75c 534 i = imgp->args->begin_envv - imgp->args->begin_argv;
19bfc8ab
MD
535 if (sizeof(struct pargs) + i <= ps_arg_cache_limit) {
536 pa = kmalloc(sizeof(struct pargs) + i, M_PARGS, M_WAITOK);
537 refcount_init(&pa->ar_ref, 1);
538 pa->ar_length = i;
539 bcopy(imgp->args->begin_argv, pa->ar_args, i);
540 KKASSERT(p->p_args == NULL);
541 p->p_args = pa;
984263bc
MD
542 }
543
544exec_fail_dealloc:
545
546 /*
547 * free various allocated resources
548 */
549 if (imgp->firstpage)
550 exec_unmap_first_page(imgp);
551
984263bc 552 if (imgp->vp) {
984263bc 553 vrele(imgp->vp);
fad57d0e 554 imgp->vp = NULL;
984263bc
MD
555 }
556
bf52a6ac
MD
557 if (imgp->freepath)
558 kfree(imgp->freepath, M_TEMP);
559
a6f89c72
DR
560 if (error == 0) {
561 ++mycpu->gd_cnt.v_exec;
19bfc8ab 562 lwkt_reltoken(&p->p_token);
984263bc 563 return (0);
a6f89c72 564 }
984263bc
MD
565
566exec_fail:
dc52e1cc
MD
567 /*
568 * we're done here, clear P_INEXEC if we were the ones that
569 * set it. Otherwise if vmspace_destroyed is still set we
570 * raced another thread and that thread is responsible for
571 * clearing it.
572 */
573 if (imgp->vmspace_destroyed & 2)
4643740a 574 p->p_flags &= ~P_INEXEC;
19bfc8ab 575 lwkt_reltoken(&p->p_token);
984263bc 576 if (imgp->vmspace_destroyed) {
5417cd57
SS
577 /*
578 * Sorry, no more process anymore. exit gracefully.
579 * However we can't die right here, because our
580 * caller might have to clean up, so indicate a
90947cb1 581 * lethal error by returning -1.
5417cd57
SS
582 */
583 return(-1);
984263bc
MD
584 } else {
585 return(error);
586 }
587}
588
2bd9d75c
DRJ
589/*
590 * execve() system call.
591 */
592int
753fd850 593sys_execve(struct execve_args *uap)
2bd9d75c 594{
fad57d0e 595 struct nlookupdata nd;
2bd9d75c
DRJ
596 struct image_args args;
597 int error;
598
118cec38 599 bzero(&args, sizeof(args));
3919ced0 600
3919ced0 601 error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW);
fad57d0e
MD
602 if (error == 0) {
603 error = exec_copyin_args(&args, uap->fname, PATH_USERSPACE,
604 uap->argv, uap->envv);
605 }
2abfe22f
MD
606 if (error == 0)
607 error = kern_execve(&nd, &args);
fad57d0e 608 nlookup_done(&nd);
2bd9d75c
DRJ
609 exec_free_args(&args);
610
5417cd57 611 if (error < 0) {
90947cb1 612 /* We hit a lethal error condition. Let's die now. */
5417cd57
SS
613 exit1(W_EXITCODE(0, SIGABRT));
614 /* NOTREACHED */
615 }
616
2bd9d75c
DRJ
617 /*
618 * The syscall result is returned in registers to the new program.
619 * Linux will register %edx as an atexit function and we must be
620 * sure to set it to 0. XXX
621 */
622 if (error == 0)
623 uap->sysmsg_result64 = 0;
624
625 return (error);
626}
627
984263bc 628int
7c34d798 629exec_map_page(struct image_params *imgp, vm_pindex_t pageno,
5c5185ae 630 struct lwbuf **plwb, const char **pdata)
984263bc 631{
1b9d3514
MD
632 int rv;
633 vm_page_t ma;
06ecca5a 634 vm_page_t m;
984263bc
MD
635 vm_object_t object;
636
fad57d0e 637 /*
7540ab49 638 * The file has to be mappable.
fad57d0e 639 */
7540ab49 640 if ((object = imgp->vp->v_object) == NULL)
fad57d0e 641 return (EIO);
984263bc 642
7c34d798
SS
643 if (pageno >= object->size)
644 return (EIO);
645
b12defdc 646 vm_object_hold(object);
7c34d798 647 m = vm_page_grab(object, pageno, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
1b9d3514
MD
648 while ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
649 ma = m;
984263bc 650
06ecca5a
MD
651 /*
652 * get_pages unbusies all the requested pages except the
a55afca2
MD
653 * primary page (at index 0 in this case). The primary
654 * page may have been wired during the pagein (e.g. by
655 * the buffer cache) so vnode_pager_freepage() must be
656 * used to properly release it.
06ecca5a 657 */
1b9d3514 658 rv = vm_pager_get_page(object, &ma, 1);
7c34d798 659 m = vm_page_lookup(object, pageno);
984263bc 660
06ecca5a
MD
661 if (rv != VM_PAGER_OK || m == NULL || m->valid == 0) {
662 if (m) {
663 vm_page_protect(m, VM_PROT_NONE);
a55afca2 664 vnode_pager_freepage(m);
984263bc 665 }
984263bc
MD
666 return EIO;
667 }
668 }
b12defdc 669 vm_page_hold(m);
06ecca5a 670 vm_page_wakeup(m); /* unbusy the page */
b12defdc 671 vm_object_drop(object);
984263bc 672
7a683a24 673 *plwb = lwbuf_alloc(m, *plwb);
5c5185ae 674 *pdata = (void *)lwbuf_kva(*plwb);
7c34d798
SS
675
676 return (0);
677}
678
7484bdd2
MD
679/*
680 * Map the first page of an executable image.
681 *
682 * NOTE: If the mapping fails we have to NULL-out firstpage which may
683 * still be pointing to our supplied lwp structure.
684 */
7c34d798
SS
685int
686exec_map_first_page(struct image_params *imgp)
687{
688 int err;
689
690 if (imgp->firstpage)
691 exec_unmap_first_page(imgp);
692
7a683a24 693 imgp->firstpage = &imgp->firstpage_cache;
7c34d798
SS
694 err = exec_map_page(imgp, 0, &imgp->firstpage, &imgp->image_header);
695
7484bdd2
MD
696 if (err) {
697 imgp->firstpage = NULL;
7c34d798 698 return err;
7484bdd2 699 }
984263bc
MD
700
701 return 0;
702}
703
704void
5c5185ae 705exec_unmap_page(struct lwbuf *lwb)
984263bc 706{
18f40545
MD
707 vm_page_t m;
708
5fd012e0 709 crit_enter();
5c5185ae
SG
710 if (lwb != NULL) {
711 m = lwbuf_page(lwb);
712 lwbuf_free(lwb);
e4ba7818 713 vm_page_unhold(m);
984263bc 714 }
5fd012e0 715 crit_exit();
984263bc
MD
716}
717
7c34d798
SS
718void
719exec_unmap_first_page(struct image_params *imgp)
720{
721 exec_unmap_page(imgp->firstpage);
722 imgp->firstpage = NULL;
723 imgp->image_header = NULL;
724}
725
984263bc
MD
726/*
727 * Destroy old address space, and allocate a new stack
728 * The new stack is only SGROWSIZ large because it is grown
729 * automatically in trap.c.
dc52e1cc
MD
730 *
731 * This is the point of no return.
984263bc
MD
732 */
733int
29802dbb 734exec_new_vmspace(struct image_params *imgp, struct vmspace *vmcopy)
984263bc 735{
984263bc
MD
736 struct vmspace *vmspace = imgp->proc->p_vmspace;
737 vm_offset_t stack_addr = USRSTACK - maxssiz;
dc52e1cc 738 struct proc *p;
29802dbb 739 vm_map_t map;
dc52e1cc 740 int error;
984263bc 741
dc52e1cc
MD
742 /*
743 * Indicate that we cannot gracefully error out any more, kill
744 * any other threads present, and set P_INEXEC to indicate that
745 * we are now messing with the process structure proper.
746 *
747 * If killalllwps() races return an error which coupled with
748 * vmspace_destroyed will cause us to exit. This is what we
749 * want since another thread is patiently waiting for us to exit
750 * in that case.
751 */
752 p = curproc;
984263bc
MD
753 imgp->vmspace_destroyed = 1;
754
dc52e1cc
MD
755 if (curthread->td_proc->p_nthreads > 1) {
756 error = killalllwps(1);
757 if (error)
758 return (error);
759 }
760 imgp->vmspace_destroyed |= 2; /* we are responsible for P_INEXEC */
4643740a 761 p->p_flags |= P_INEXEC;
08f2f1bb
SS
762
763 /*
984263bc
MD
764 * Blow away entire process VM, if address space not shared,
765 * otherwise, create a new VM space so that other threads are
29802dbb
MD
766 * not disrupted. If we are execing a resident vmspace we
767 * create a duplicate of it and remap the stack.
239b4df9
MD
768 *
769 * The exitingcnt test is not strictly necessary but has been
770 * included for code sanity (to make the code more deterministic).
984263bc 771 */
29802dbb
MD
772 map = &vmspace->vm_map;
773 if (vmcopy) {
774 vmspace_exec(imgp->proc, vmcopy);
775 vmspace = imgp->proc->p_vmspace;
776 pmap_remove_pages(vmspace_pmap(vmspace), stack_addr, USRSTACK);
777 map = &vmspace->vm_map;
e3161323
MD
778 } else if (vmspace->vm_sysref.refcnt == 1 &&
779 vmspace->vm_exitingcnt == 0) {
fef0fdf2 780 shmexit(vmspace);
239b4df9 781 if (vmspace->vm_upcalls)
08f2f1bb 782 upc_release(vmspace, ONLY_LWP_IN_PROC(imgp->proc));
88181b08 783 pmap_remove_pages(vmspace_pmap(vmspace),
b0c15cdf 784 0, VM_MAX_USER_ADDRESS);
88181b08 785 vm_map_remove(map, 0, VM_MAX_USER_ADDRESS);
984263bc 786 } else {
29802dbb 787 vmspace_exec(imgp->proc, NULL);
984263bc
MD
788 vmspace = imgp->proc->p_vmspace;
789 map = &vmspace->vm_map;
790 }
791
792 /* Allocate a new stack */
793 error = vm_map_stack(&vmspace->vm_map, stack_addr, (vm_size_t)maxssiz,
c809941b 794 0, VM_PROT_ALL, VM_PROT_ALL, 0);
984263bc
MD
795 if (error)
796 return (error);
797
798 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
799 * VM_STACK case, but they are still used to monitor the size of the
800 * process stack so we can check the stack rlimit.
801 */
802 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
803 vmspace->vm_maxsaddr = (char *)USRSTACK - maxssiz;
804
805 return(0);
806}
807
808/*
809 * Copy out argument and environment strings from the old process
810 * address space into the temporary string buffer.
811 */
812int
2bd9d75c 813exec_copyin_args(struct image_args *args, char *fname,
2abfe22f 814 enum exec_path_segflg segflg, char **argv, char **envv)
984263bc 815{
984263bc 816 char *argp, *envp;
2bd9d75c 817 int error = 0;
984263bc
MD
818 size_t length;
819
18b3f457 820 args->buf = objcache_get(exec_objcache, M_WAITOK);
2bd9d75c
DRJ
821 if (args->buf == NULL)
822 return (ENOMEM);
823 args->begin_argv = args->buf;
824 args->endp = args->begin_argv;
825 args->space = ARG_MAX;
2bd9d75c
DRJ
826
827 args->fname = args->buf + ARG_MAX;
828
984263bc 829 /*
2bd9d75c 830 * Copy the file name.
984263bc 831 */
2bd9d75c
DRJ
832 if (segflg == PATH_SYSSPACE) {
833 error = copystr(fname, args->fname, PATH_MAX, &length);
834 } else if (segflg == PATH_USERSPACE) {
835 error = copyinstr(fname, args->fname, PATH_MAX, &length);
836 }
984263bc 837
2bd9d75c 838 /*
66be6566
MD
839 * Extract argument strings. argv may not be NULL. The argv
840 * array is terminated by a NULL entry. We special-case the
841 * situation where argv[0] is NULL by passing { filename, NULL }
842 * to the new program to guarentee that the interpreter knows what
843 * file to open in case we exec an interpreted file. Note that
844 * a NULL argv[0] terminates the argv[] array.
845 *
846 * XXX the special-casing of argv[0] is historical and needs to be
847 * revisited.
2bd9d75c 848 */
66be6566
MD
849 if (argv == NULL)
850 error = EFAULT;
851 if (error == 0) {
852 while ((argp = (caddr_t)(intptr_t)fuword(argv++)) != NULL) {
853 if (argp == (caddr_t)-1) {
2bd9d75c 854 error = EFAULT;
66be6566 855 break;
2bd9d75c
DRJ
856 }
857 error = copyinstr(argp, args->endp,
2abfe22f 858 args->space, &length);
66be6566
MD
859 if (error) {
860 if (error == ENAMETOOLONG)
861 error = E2BIG;
862 break;
863 }
2bd9d75c
DRJ
864 args->space -= length;
865 args->endp += length;
866 args->argc++;
984263bc 867 }
66be6566
MD
868 if (args->argc == 0 && error == 0) {
869 length = strlen(args->fname) + 1;
870 if (length > args->space) {
871 error = E2BIG;
872 } else {
873 bcopy(args->fname, args->endp, length);
874 args->space -= length;
875 args->endp += length;
876 args->argc++;
877 }
878 }
984263bc
MD
879 }
880
2bd9d75c 881 args->begin_envv = args->endp;
984263bc
MD
882
883 /*
66be6566 884 * extract environment strings. envv may be NULL.
984263bc 885 */
2abfe22f 886 if (envv && error == 0) {
984263bc 887 while ((envp = (caddr_t) (intptr_t) fuword(envv++))) {
2bd9d75c
DRJ
888 if (envp == (caddr_t) -1) {
889 error = EFAULT;
66be6566 890 break;
984263bc 891 }
2bd9d75c
DRJ
892 error = copyinstr(envp, args->endp, args->space,
893 &length);
66be6566
MD
894 if (error) {
895 if (error == ENAMETOOLONG)
896 error = E2BIG;
897 break;
898 }
2bd9d75c
DRJ
899 args->space -= length;
900 args->endp += length;
901 args->envc++;
984263bc
MD
902 }
903 }
2bd9d75c
DRJ
904 return (error);
905}
906
907void
908exec_free_args(struct image_args *args)
909{
2abfe22f 910 if (args->buf) {
18b3f457 911 objcache_put(exec_objcache, args->buf);
2abfe22f
MD
912 args->buf = NULL;
913 }
984263bc
MD
914}
915
916/*
917 * Copy strings out to the new process address space, constructing
918 * new arg and env vector tables. Return a pointer to the base
919 * so that it can be used as the initial stack pointer.
920 */
921register_t *
2bd9d75c 922exec_copyout_strings(struct image_params *imgp)
984263bc 923{
17a7ca1f 924 int argc, envc, sgap;
984263bc
MD
925 char **vectp;
926 char *stringp, *destp;
927 register_t *stack_base;
928 struct ps_strings *arginfo;
adc42cf3 929 size_t execpath_len;
984263bc
MD
930 int szsigcode;
931
932 /*
933 * Calculate string base and vector table pointers.
934 * Also deal with signal trampoline code for this exec type.
935 */
adc42cf3
JM
936 if (imgp->execpath != NULL && imgp->auxargs != NULL)
937 execpath_len = strlen(imgp->execpath) + 1;
938 else
939 execpath_len = 0;
984263bc
MD
940 arginfo = (struct ps_strings *)PS_STRINGS;
941 szsigcode = *(imgp->proc->p_sysent->sv_szsigcode);
17a7ca1f 942 if (stackgap_random != 0)
0ced1954 943 sgap = ALIGN(karc4random() & (stackgap_random - 1));
17a7ca1f
MD
944 else
945 sgap = 0;
946 destp = (caddr_t)arginfo - szsigcode - SPARE_USRSPACE - sgap -
adc42cf3 947 roundup(execpath_len, sizeof(char *)) -
2bd9d75c 948 roundup((ARG_MAX - imgp->args->space), sizeof(char *));
984263bc
MD
949
950 /*
951 * install sigcode
952 */
953 if (szsigcode)
954 copyout(imgp->proc->p_sysent->sv_sigcode,
2bd9d75c 955 ((caddr_t)arginfo - szsigcode), szsigcode);
984263bc
MD
956
957 /*
adc42cf3
JM
958 * Copy the image path for the rtld
959 */
960 if (execpath_len != 0) {
961 imgp->execpathp = (uintptr_t)arginfo
962 - szsigcode
963 - execpath_len;
964 copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len);
965 }
966
967 /*
984263bc
MD
968 * If we have a valid auxargs ptr, prepare some room
969 * on the stack.
7ec1918b 970 *
984263bc
MD
971 * The '+ 2' is for the null pointers at the end of each of the
972 * arg and env vector sets, and 'AT_COUNT*2' is room for the
973 * ELF Auxargs data.
974 */
7ec1918b
MD
975 if (imgp->auxargs) {
976 vectp = (char **)(destp - (imgp->args->argc +
adc42cf3
JM
977 imgp->args->envc + 2 + (AT_COUNT * 2) + execpath_len) *
978 sizeof(char*));
7ec1918b 979 } else {
2bd9d75c 980 vectp = (char **)(destp - (imgp->args->argc +
7ec1918b
MD
981 imgp->args->envc + 2) * sizeof(char*));
982 }
983
984263bc 984 /*
31bc42c8
MD
985 * NOTE: don't bother aligning the stack here for GCC 2.x, it will
986 * be done in crt1.o. Note that GCC 3.x aligns the stack in main.
984263bc 987 */
984263bc
MD
988
989 /*
990 * vectp also becomes our initial stack base
991 */
992 stack_base = (register_t *)vectp;
993
2bd9d75c
DRJ
994 stringp = imgp->args->begin_argv;
995 argc = imgp->args->argc;
996 envc = imgp->args->envc;
984263bc
MD
997
998 /*
999 * Copy out strings - arguments and environment.
1000 */
2bd9d75c 1001 copyout(stringp, destp, ARG_MAX - imgp->args->space);
984263bc
MD
1002
1003 /*
1004 * Fill in "ps_strings" struct for ps, w, etc.
1005 */
1006 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1007 suword(&arginfo->ps_nargvstr, argc);
1008
1009 /*
1010 * Fill in argument portion of vector table.
1011 */
1012 for (; argc > 0; --argc) {
1013 suword(vectp++, (long)(intptr_t)destp);
1014 while (*stringp++ != 0)
1015 destp++;
1016 destp++;
1017 }
1018
7a44d1cb 1019 /* a null vector table pointer separates the argp's from the envp's */
984263bc
MD
1020 suword(vectp++, 0);
1021
1022 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1023 suword(&arginfo->ps_nenvstr, envc);
1024
1025 /*
1026 * Fill in environment portion of vector table.
1027 */
1028 for (; envc > 0; --envc) {
1029 suword(vectp++, (long)(intptr_t)destp);
1030 while (*stringp++ != 0)
1031 destp++;
1032 destp++;
1033 }
1034
1035 /* end of vector table is a null pointer */
1036 suword(vectp, 0);
1037
1038 return (stack_base);
1039}
1040
1041/*
1042 * Check permissions of file to execute.
1043 * Return 0 for success or error code on failure.
1044 */
1045int
246693ac 1046exec_check_permissions(struct image_params *imgp, struct mount *topmnt)
984263bc
MD
1047{
1048 struct proc *p = imgp->proc;
1049 struct vnode *vp = imgp->vp;
1050 struct vattr *attr = imgp->attr;
1051 int error;
1052
1053 /* Get file attributes */
87de5057 1054 error = VOP_GETATTR(vp, attr);
984263bc
MD
1055 if (error)
1056 return (error);
1057
1058 /*
1059 * 1) Check if file execution is disabled for the filesystem that this
1060 * file resides on.
1061 * 2) Insure that at least one execute bit is on - otherwise root
1062 * will always succeed, and we don't want to happen unless the
1063 * file really is executable.
1064 * 3) Insure that the file is a regular file.
1065 */
1066 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
246693ac 1067 ((topmnt != NULL) && (topmnt->mnt_flag & MNT_NOEXEC)) ||
984263bc
MD
1068 ((attr->va_mode & 0111) == 0) ||
1069 (attr->va_type != VREG)) {
1070 return (EACCES);
1071 }
1072
1073 /*
1074 * Zero length files can't be exec'd
1075 */
1076 if (attr->va_size == 0)
1077 return (ENOEXEC);
1078
1079 /*
1080 * Check for execute permission to file based on current credentials.
1081 */
90a83564 1082 error = VOP_EACCESS(vp, VEXEC, p->p_ucred);
984263bc
MD
1083 if (error)
1084 return (error);
1085
1086 /*
1087 * Check number of open-for-writes on the file and deny execution
1088 * if there are any.
1089 */
1090 if (vp->v_writecount)
1091 return (ETXTBSY);
1092
1093 /*
7540ab49
MD
1094 * Call filesystem specific open routine, which allows us to read,
1095 * write, and mmap the file. Without the VOP_OPEN we can only
1096 * stat the file.
984263bc 1097 */
87de5057 1098 error = VOP_OPEN(vp, FREAD, p->p_ucred, NULL);
984263bc
MD
1099 if (error)
1100 return (error);
1101
1102 return (0);
1103}
1104
1105/*
1106 * Exec handler registration
1107 */
1108int
77153250 1109exec_register(const struct execsw *execsw_arg)
984263bc
MD
1110{
1111 const struct execsw **es, **xs, **newexecsw;
1112 int count = 2; /* New slot and trailing NULL */
1113
1114 if (execsw)
1115 for (es = execsw; *es; es++)
1116 count++;
efda3bd0 1117 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK);
984263bc
MD
1118 xs = newexecsw;
1119 if (execsw)
1120 for (es = execsw; *es; es++)
1121 *xs++ = *es;
1122 *xs++ = execsw_arg;
1123 *xs = NULL;
1124 if (execsw)
efda3bd0 1125 kfree(execsw, M_TEMP);
984263bc
MD
1126 execsw = newexecsw;
1127 return 0;
1128}
1129
1130int
77153250 1131exec_unregister(const struct execsw *execsw_arg)
984263bc
MD
1132{
1133 const struct execsw **es, **xs, **newexecsw;
1134 int count = 1;
1135
1136 if (execsw == NULL)
ae7e74d4 1137 panic("unregister with no handlers left?");
984263bc
MD
1138
1139 for (es = execsw; *es; es++) {
1140 if (*es == execsw_arg)
1141 break;
1142 }
1143 if (*es == NULL)
1144 return ENOENT;
1145 for (es = execsw; *es; es++)
1146 if (*es != execsw_arg)
1147 count++;
efda3bd0 1148 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK);
984263bc
MD
1149 xs = newexecsw;
1150 for (es = execsw; *es; es++)
1151 if (*es != execsw_arg)
1152 *xs++ = *es;
1153 *xs = NULL;
1154 if (execsw)
efda3bd0 1155 kfree(execsw, M_TEMP);
984263bc
MD
1156 execsw = newexecsw;
1157 return 0;
1158}